If non-standard tax questions are taking too long to research, your firm is not alone and does not need to adopt an unfamiliar system blindly. This guide explains what AI-assisted tax research can and cannot do, compares five products worth investigating, and gives you a practical way to build a shortlist.
In short: Blue J is our provisional top pick for a US CPA firm seeking a focused AI tax research experience, while CoCounsel Tax deserves first consideration for firms already invested in the Thomson Reuters research environment. CCH AnswerConnect and Bloomberg Tax AI Assistant are ecosystem-led alternatives, while TaxGPT may appeal to firms seeking a more focused standalone option. Verify current coverage, pricing, security terms and integrations before signing.
NTK Score: Blue J
Our editorial assessment of how confidently we'd recommend this to a small business, weighing real-world usability, commercial value, effort, long-term risk and practical experience — not a popularity score or a compliance audit.
Blue J tightly matches US tax research through federal and all-state coverage, cited answers, file analysis, and drafting, but current materials emphasize research rather than the legacy prediction engine.
Tier 2 · Vendor pricing and coverage: https://www.bluej.com/pricing and https://support.bluej.com/hc/en-us/articles/30133433035540-What-data-is-Blue-J-trained-on; Tier 3 practitioner signal: https://www.reddit.com/r/taxpros/comments/1l8jdne/bluej_ai_tax_research/ · Confidence: Moderate
Browser-based setup, a no-card trial, clear tutorials, and copy, PDF, and DOCX output reduce implementation work, while undocumented practice-management integrations and sales-led team controls limit plug-and-play deployment.
Tier 2 · Vendor pricing, onboarding, and output documentation: https://www.bluej.com/pricing, https://support.bluej.com/hc/en-us/articles/30131993170068-Signing-in-to-your-account, and https://support.bluej.com/hc/en-us/articles/30133265755540-Copying-and-printing-answers · Confidence: Low
Conversational workflows, source highlighting, follow-up prompts, and training aid adoption, but tax professionals must learn effective prompting and independently verify answers that practitioners report can still be wrong.
Tier 2 · Vendor usage documentation: https://support.bluej.com/hc/en-us/articles/30133164965524-Ask-your-first-question-in-Blue-J and https://support.bluej.com/hc/en-us/articles/30133249231124-How-to-view-and-use-sources-in-Blue-J; Tier 3 practitioner signal: https://www.reddit.com/r/taxpros/comments/1l8jdne/bluej_ai_tax_research/ · Confidence: Moderate
The $1,498 annual individual plan includes unlimited use and a no-card trial, offering credible ROI for research-heavy firms, but team pricing, renewal exposure, and some secondary-source access remain opaque or extra.
Tier 2 · Vendor pricing and contract terms: https://www.bluej.com/pricing and https://www.bluej.com/subscriber-agreement; source-access documentation: https://support.bluej.com/hc/en-us/articles/30133249231124-How-to-view-and-use-sources-in-Blue-J; CPA.com partner pricing: https://www.cpa.com/tax-research; Tier 3 pricing and value signal: https://www.reddit.com/r/taxpros/comments/1cx7qlb/has_anyone_used_blue_j_ai_in_tax/ · Confidence: Moderate
The vendor states it maintains SOC 2 Type 2, US-only residency, encryption, logging, and no model training, but broad contractual input-review rights and plan-dependent access controls warrant diligence.
Tier 2 · Vendor security, contractual, and funding disclosures: https://www.bluej.com/security, https://www.bluej.com/terms-of-use, https://www.bluej.com/subscriber-agreement, and https://www.bluej.com/ca/blog/blue-j-announces-122m-series-d-financing-led-by-oak-hc-ft-and-sapphire-ventures; no reliable independent security or support benchmark was found · Confidence: Low
What to know before you buy
These don't change the score above — they're conditions worth understanding for specific readers before you commit.
The current Terms of Use grants broad, perpetual rights over submitted data for platform improvement, and the Subscriber Agreement permits review of all inputs and outputs; the security page separately says customer data is not used to train generative models. Sources: https://www.bluej.com/terms-of-use, https://www.bluej.com/subscriber-agreement, and https://www.bluej.com/security
Who this matters to: US tax firms planning to upload unredacted returns, Social Security numbers, or other confidential client records — The contractual permission for platform improvement and input review may conflict with a firm's client-confidentiality, data-minimization, or internal AI-governance requirements despite the no-model-training statement.
What to do: Do not upload unredacted client files until a signed order or data-processing agreement defines human access, permitted uses, deletion, retention, and subprocessor handling to the firm's satisfaction.
Changes our recommendation for this audience — see verdict notes.
Blue J's About page says the company began with a tax-law prediction engine but launched generative tax research in 2023 as its flagship; current pricing lists research, drafting, citations, and uploads without a prediction feature. Sources: https://www.bluej.com/uk/about-us and https://www.bluej.com/pricing
Who this matters to: Firms specifically purchasing software to predict tax-case or dispute outcomes — A buyer seeking outcome prediction could purchase a research-focused product that does not provide the expected predictive workflow.
What to do: Require a live demonstration and written confirmation of current US prediction functionality before purchasing for that use case.
Changes our recommendation for this audience — see verdict notes.
Blue J's help documentation says viewing full IBFD-sourced international content and Tax Notes commentary requires separate subscriptions, although citations and excerpts may appear in Blue J. Source: https://support.bluej.com/hc/en-us/articles/30133249231124-How-to-view-and-use-sources-in-Blue-J
Who this matters to: Firms expecting the Blue J fee to include unrestricted access to every cited secondary source — Researchers may encounter paywalled source text, increasing total cost or limiting full verification of secondary commentary.
What to do: Audit likely source requirements during the trial and price any necessary Tax Notes or IBFD subscriptions into the purchase decision.
The standard Subscriber Agreement provides automatic renewal unless written notice is given at least 14 days before expiry and allows remaining subscription fees to be accelerated after some terminations. Source: https://www.bluej.com/subscriber-agreement
Who this matters to: Small firms with seasonal cash flow or uncertain long-term usage — Missing the notice window may create another annual commitment, while early termination may not eliminate remaining payment exposure.
What to do: Review the final order for overrides, calendar the cancellation deadline, and obtain any negotiated termination terms in writing.
Blue J lists SSO and custom access controls under the quote-priced Team plan, while its MFA guide says only some accounts require MFA. Sources: https://www.bluej.com/pricing and https://support.bluej.com/hc/en-us/articles/30133047451028-Setting-Up-Multi-Factor-Authentication-MFA
Who this matters to: Firms requiring mandatory MFA, centralized identity management, or custom user controls on every account — The individual plan may not satisfy the firm's internal access-control policy, and the compliant team configuration has undisclosed pricing.
What to do: Confirm mandatory MFA, administrator controls, SSO availability, audit-log access, and total team pricing before procurement.
Recommendation depends on who's asking
US tax firms planning to upload unredacted returns, Social Security numbers, or other confidential client records: The contractual permission for platform improvement and input review may conflict with a firm's client-confidentiality, data-minimization, or internal AI-governance requirements despite the no-model-training statement.
Firms specifically purchasing software to predict tax-case or dispute outcomes: A buyer seeking outcome prediction could purchase a research-focused product that does not provide the expected predictive workflow.
What an AI tax research tool is meant to change
The practical goal is not to let software make tax decisions. It is to reduce the time between receiving an unfamiliar question and assembling a defensible research trail for review by a qualified professional.
Consider the named reader for this guide: a tax manager at a US CPA firm researching a multi-state filing question under deadline pressure. Before adopting a suitable tool, the manager might search IRS material, state sources, editorial explanations and case law separately, then reconcile the results manually. After adoption, the tool should help surface potentially relevant authority and citations in minutes, while the preparer opens every cited source, checks its current status and documents the final reasoning.
That distinction matters. A fluent answer is not the work product. The reviewed authorities, reasoning and documented conclusion are the work product.
Tax prompts can contain taxpayer identities, transaction details and other confidential information. Before staff enter real client facts, confirm what the vendor stores, who can access it, whether prompts train models, how long data is retained and which contractual terms apply to your firm.
How we evaluated the shortlist
This guide compares the five named products against the factors that matter in a professional tax research workflow. Because pricing and product details can change, the shortlist should start a structured trial rather than replace current vendor documentation.
The most important criteria are:
- US authority coverage: The system should make it clear which federal and state materials it searches, including any gaps in historical, administrative or judicial sources.
- Citation traceability: A reviewer should be able to move from an answer to the underlying authority without reconstructing the search from scratch.
- Source status and context: Citations need enough context for a tax professional to assess jurisdiction, date, hierarchy and subsequent treatment.
- Answer discipline: A useful product distinguishes sourced statements from synthesis and makes uncertainty visible.
- Review workflow: Saving research, sharing it with another preparer and recording reviewer approval should be straightforward.
- Data handling: The contract and security materials should address prompts, uploaded documents, retention, model training, subprocessors and incident response.
- Firm fit: Existing research subscriptions, identity management and document systems can make an ecosystem product more practical than a theoretically stronger standalone tool.
- Team-level cost: Compare the complete annual cost for researchers, reviewers, implementation and overlapping subscriptions, not an attractive per-user figure in isolation.
No product earns credit merely for producing the fastest answer. In tax research, a slower answer with direct, reviewable support can be more valuable than a polished response whose reasoning cannot be reconstructed.
Comparison: five AI tax research tools
Pricing is intentionally shown as unverified rather than guessed. As of July 2026, this offline draft could not independently confirm current US pricing, packaging or contract terms. Ask each vendor for a written quote covering the whole research team and any required underlying subscription.
| Tool | Best initial fit | What may distinguish it | Pricing status | Main point to verify |
|---|---|---|---|---|
| Blue J | Firms wanting a focused AI tax research workflow | Research-first experience with cited responses | Obtain and verify a current USD quote | Exact federal and state source coverage, citation links and export workflow |
| CoCounsel Tax | Firms already using Thomson Reuters tax research products | Potential value from its wider professional research ecosystem | Obtain and verify a current USD quote | Required subscriptions, included content and workflow integrations |
| TaxGPT | Firms considering a focused standalone tax assistant | Potentially simpler route to AI-assisted tax questions | Obtain and verify a current USD quote | Authority depth, state coverage, security terms and administrative controls |
| CCH AnswerConnect | Firms already standardised on Wolters Kluwer research | Potential continuity with existing tax research content | Obtain and verify a current USD quote | Which AI functions are currently available and what content they search |
| Bloomberg Tax AI Assistant | Firms whose research already centres on Bloomberg Tax | Potential value from the Bloomberg Tax content environment | Obtain and verify a current USD quote | Availability, entitlements, supported sources and collaboration features |
This table describes the buying position each product appears likely to occupy, not a verified feature inventory. A firm should require a live demonstration using its own anonymised research questions before drawing conclusions.
Blue J: provisional best overall starting point
Blue J is the strongest first demonstration for a firm that wants AI-assisted tax research to be the main workflow rather than an extra feature inside a broader platform. Its apparent research-first positioning makes it easier to evaluate against the central question: can the product move a tax professional from a fact pattern to relevant, traceable authority without obscuring uncertainty?
The recommendation is provisional because current source coverage, product packaging and data terms require primary-source verification. During a demonstration, ask Blue J to answer the same federal and multi-state questions your team gives every contender. Have a senior reviewer score whether the response identifies controlling authority, separates jurisdictions and provides usable links.
Potential advantages
- A focused product may be easier to trial against a defined tax research workflow.
- It appears positioned around sourced tax answers rather than general office assistance.
- It may suit firms that do not want the choice dictated by an existing publishing subscription.
Limitations to investigate
- State-by-state depth may vary and must be tested explicitly.
- A focused interface does not automatically provide the document, workflow or identity integrations a larger platform can offer.
- The full-team cost and any underlying content dependencies need written confirmation.
Verdict: Put Blue J first on the demonstration schedule, but do not approve it until your reviewers have validated its sources on representative questions and your security lead has reviewed the contract.
CoCounsel Tax: best ecosystem candidate for Thomson Reuters firms
CoCounsel Tax should be near the top of the list when a firm already relies on Thomson Reuters tax research or related professional products. The practical attraction is not simply an AI answer. It is the possibility of keeping research content and AI-assisted work within a familiar vendor environment.
That ecosystem advantage can reduce implementation friction, but only if the required content, permissions and integrations are included in the proposed package. Ask the vendor to distinguish clearly between CoCounsel Tax capabilities, underlying research subscriptions and features that require separate products.
Potential advantages
- It may reduce context switching for firms already using the relevant Thomson Reuters environment.
- Existing vendor review, procurement and account relationships may simplify adoption.
- Familiar research content could make citation review easier for current users.
Limitations to investigate
- The useful configuration may depend on other subscriptions or negotiated entitlements.
- Firms outside the ecosystem may pay for breadth they do not need.
- Product naming and packaging can change, so the proposal must identify every included component.
Verdict: For an established Thomson Reuters customer, compare CoCounsel Tax directly with Blue J before considering a platform migration.
TaxGPT: a focused option that needs a rigorous authority test
TaxGPT may appeal to firms that want a dedicated tax question-and-answer product without adopting a large research ecosystem. That narrower proposition could make a pilot easier to explain to staff, but simplicity at the interface should not be confused with depth behind the answer.
Use the trial to examine hard cases, not common questions that any system can answer convincingly. Include questions with conflicting facts, multiple jurisdictions, superseded guidance and an intentionally false premise. A dependable research assistant should expose ambiguity rather than smooth it away.
Potential advantages
- A focused interface may make initial staff training easier.
- It could suit a small research group seeking a contained pilot.
- It provides another competitive option outside the major publishing ecosystems.
Limitations to investigate
- Confirm exactly which primary and secondary sources are covered.
- Examine whether citations open to the precise passage or merely identify a document.
- Review administrative controls, retention, model-training terms and export options.
Verdict: Include TaxGPT when ease of entry matters, but make source quality and security documentation pass-or-fail requirements.
CCH AnswerConnect: best candidate for existing Wolters Kluwer users
CCH AnswerConnect belongs on the shortlist for a firm whose researchers already work within Wolters Kluwer tax content. The potential benefit is continuity. Staff may be able to evaluate AI-assisted results against sources and editorial materials they already understand.
The central buying question is which AI capabilities are included in the current product and which content those capabilities can search. Do not accept a demonstration that blends multiple products without explaining the licensing boundaries.
Potential advantages
- Existing users may face less change in their research habits.
- Familiar content organisation can help reviewers assess generated results.
- Consolidating with an incumbent vendor may simplify support and billing.
Limitations to investigate
- Current AI functionality and availability need direct confirmation.
- Valuable functions may sit in a particular package or require additional entitlements.
- Staying with an incumbent can reduce the pressure to test whether another workflow is materially better.
Verdict: Existing CCH customers should test AnswerConnect before adding a separate vendor, but renewal convenience should not replace a controlled comparison.
Bloomberg Tax AI Assistant: best candidate for Bloomberg-centred research teams
Bloomberg Tax AI Assistant is the logical fifth contender for firms already using Bloomberg Tax as a major research source. Its potential value depends on whether AI-assisted research works across the content and practice areas the firm actually licenses.
Ask for an account-specific demonstration rather than a generic sales environment. The evaluator should confirm which sources are available to ordinary users, whether answers preserve links to underlying material and how research can be saved or shared.
Potential advantages
- It may fit teams already comfortable with Bloomberg Tax content.
- Keeping research in an existing environment can reduce switching between tools.
- A current vendor relationship may simplify contracting and support escalation.
Limitations to investigate
- Availability and entitlements may depend on the firm's subscription.
- The firm should verify whether the assistant covers every needed federal and state workflow.
- It may be difficult to separate the value of the assistant from the cost of the broader subscription.
Verdict: Bloomberg Tax customers should evaluate the assistant as an extension of their current research stack, while non-customers should first decide whether the underlying content platform belongs on their shortlist.
Why Blue J is the top pick
Blue J is our provisional top pick because a focused research product gives a US CPA firm the clearest starting point for testing the core workflow. It can be compared directly against the firm's existing manual process without first justifying a broad platform change.
The choice changes when an incumbent ecosystem already works well. A Thomson Reuters firm should give CoCounsel Tax equal consideration, while established Wolters Kluwer or Bloomberg Tax customers should test the corresponding option before buying overlapping content.
There is also a valid case for buying nothing yet. If reviewers cannot open citations quickly, cannot establish which authority supports each material statement or cannot keep sensitive client facts out of inappropriate data flows, improved research templates and stronger internal review may be the better immediate investment.
What US firms should check before choosing
Build a representative question set
Select 15 to 25 anonymised questions from completed work. Include federal questions, multi-state matters, uncommon fact patterns, old authority, conflicting authority and questions whose correct response is that more facts are required.
Score every product against the same answer key. Record missed authorities, irrelevant citations, invented citations, unsupported conclusions and the time a reviewer needs to reach a usable workpaper.
Test citation quality, not citation quantity
A long list of references can create false confidence. Reviewers should be able to identify the relevant passage, jurisdiction, effective period and authority level, then determine whether the material remains applicable.
A product that cites ten loosely related documents may be less useful than one that presents three directly relevant sources with clear context. Require reviewers to open every material citation during the pilot.
Review confidentiality and data terms
The FTC's Safeguards Rule guidance expressly lists tax preparation firms among the financial institutions that may be covered and says covered firms must maintain a written information security program and oversee service providers. IRS Publication 4557 says tax return preparers must create and enact security plans, while the IRS's section 7216 guidance addresses unauthorized use or disclosure of tax return information. A CPA firm should have qualified advisers apply these sources, relevant state privacy laws and professional confidentiality standards to its circumstances before client data is entered into an AI service.
Ask for the privacy notice, data-processing terms, security documentation and subprocessor list. Confirm prompt retention, uploaded-file handling, model-training restrictions, user deletion, single sign-on, access logs, incident notification and the location of relevant processing.
Price the team, not the seat
Request a written USD quote for the complete team, including tax researchers, reviewers, administrators, onboarding and any required content subscription. Compare the annual cost with the hours currently spent finding sources and checking junior research, but do not assume every saved search minute becomes billable capacity.
For example, the useful comparison is not one advertised seat against one employee's hourly rate. It is the full annual contract against measurable reductions in research and review time across the firm, adjusted for training, duplicate subscriptions and quality-control work.
Run a controlled pilot
Begin with anonymised or synthetic facts unless the contract and internal approval process permit real client information. Use a small group containing at least one experienced tax reviewer, one frequent researcher and one person responsible for security or risk.
Define success before the pilot starts. Useful measures include time to relevant authority, percentage of material citations verified, reviewer correction time, number of unsupported propositions and user adoption across repeated assignments.
Methodology (Real-World, Verified)
We score AI tools against real SMB workflows using named vendor documentation, pricing pages, and independent sources, not enterprise demos. Pricing is verified at the vendor's published rates, with local-currency conversions noted where relevant. Compliance notes reference the legislation and regulatory guidance relevant to each article's region. Every tool is judged on one question: could a business with no dedicated IT department actually pick this up and use it on Monday morning.
Read our full methodology and independence and disclosure policy.
Related reading: our AI governance by region.
Can AI tax research replace a tax professional?
No. It can accelerate searching, summarising and organising potential sources, but a qualified professional still needs to verify authorities, establish that they apply to the facts and approve the conclusion. Treat generated text as a research lead, not tax advice or a finished workpaper.
Which AI tax research tool is best for a small US CPA firm?
Blue J is the best first product to demonstrate when the firm wants a focused research workflow and has no overriding platform commitment. Firms already invested in Thomson Reuters, Wolters Kluwer or Bloomberg Tax should compare the corresponding ecosystem option before adding another subscription.
How should a firm test citation accuracy?
Use completed, anonymised matters with a reviewer-approved research trail. Require the evaluator to open each material citation, confirm the relevant passage and status, and record any unsupported or misleading statements. Common textbook questions are not a sufficient test.
Is it safe to enter client information into an AI tax tool?
Not by default. Safety depends on the data, vendor contract, technical controls and the firm's own obligations. Start with anonymised facts, then obtain internal approval only after reviewing retention, model training, access, deletion and incident terms.
Should a firm buy a standalone tool or use its existing research vendor?
Choose the option that produces the best reviewed result with the least operational friction. An incumbent platform may reduce training and integration work, while a focused standalone product may provide a better research experience. A side-by-side pilot is more reliable than deciding from feature lists.
What should be included in an AI tax research pilot?
Include representative questions, a documented answer key, citation checks, privacy review, reviewer-time measurements and clear failure thresholds. The pilot should also test how research is saved, shared, corrected and approved, not just how quickly the first answer appears.
Methodology
Need to Know AI evaluates tools against real small and medium business workflows rather than enterprise demonstrations. For this guide, the recommended evaluation uses repeated tax questions, citation verification, reviewer effort, data handling and team-level cost.
Need to Know AI checked the publicly documented product, pricing and security claims against primary vendor sources on 19 August 2026. Quote-only pricing, negotiated entitlements, contract-specific data terms and hands-on performance remain unverified and should be confirmed directly with each vendor before purchase.
Take the next step
Turn two or three contenders into a controlled shortlist. Request written coverage, security and pricing information, then run the same anonymised research set through each product before approving a purchase.
Once you've shortlisted a tax research tool, put its answers through a standard review step before anything reaches a client file.
AI Output Review Checklist for FirmsNTK Score: CoCounsel Tax
Our editorial assessment of how confidently we'd recommend this to a small business, weighing real-world usability, commercial value, effort, long-term risk and practical experience — not a popularity score or a compliance audit.
Checkpoint-grounded research, multistate coverage, document analysis, and citation-linked outputs closely fit US tax-firm work, while hallucination risk and required professional verification prevent an excellent score.
Tier 2 · Tier 2 vendor product and plan pages: https://tax.thomsonreuters.com/en/products/cocounsel-tax and https://tax.thomsonreuters.com/en/products/cocounsel-tax/plans ; Tier 3 practitioner signal: https://www.reddit.com/r/taxpros/comments/1uf6936/cocounsel_and_ready_to_advise/ and https://www.reddit.com/r/taxpros/comments/1mqivie/thomson_reuters_cocounsel_tax/ · Confidence: Moderate
Natural-language queries, templates, uploads, exports, knowledge-base integrations, and administration tools suggest a light start, but integration and migration effort is less certain outside the Thomson Reuters ecosystem.
Tier 2 · Tier 2 vendor documentation: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/workspaces/workspaces and https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/knowledge-base/knowledge-base.html ; Tier 3 practitioner workflow signal: https://www.reddit.com/r/taxpros/comments/1mqivie/thomson_reuters_cocounsel_tax/ · Confidence: Moderate
Vendor training, weekday phone and chat support, citations, and review workflows reduce onboarding burden, while staff still need prompting discipline and source-checking because practitioners report occasional hallucinations.
Tier 2 · Tier 2 vendor pricing, support, and workflow documentation: https://tax.thomsonreuters.com/en-us/products/cocounsel-tax/plans-pricing and https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/workspaces/manage-your-workspace ; Tier 3 practitioner signal: https://www.reddit.com/r/taxpros/comments/1uf6936/cocounsel_and_ready_to_advise/ · Confidence: Moderate
The vendor publishes a $1,300-per-user annual Essentials price, but caps Checkpoint-backed answers at 20 weekly and leaves full Tax Research pricing quote-only, weakening predictability and scale economics.
Tier 2 · Tier 2 vendor pricing and plan comparison: https://tax.thomsonreuters.com/en-us/products/cocounsel-tax/plans-pricing and https://tax.thomsonreuters.com/en/products/cocounsel-tax/plans ; Tier 3 price and value signal: https://www.reddit.com/r/taxpros/comments/1mqivie/thomson_reuters_cocounsel_tax/ and https://www.reddit.com/r/taxpros/comments/1s1gbrn/claude_gemini_pro_or_chatgpt_for_tax_research_and/ · Confidence: Moderate
Against NIST AI RMF expectations, vendor-stated SOC 2 Type II, encryption, US hosting, 2FA, roles, review controls, ISO 42001 inclusion, and strong finances score well, but audit-log depth and support reputation remain unclear.
Tier 2 · Tier 2 vendor security, administration, certification, and financial disclosures: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/security/security-info-cocounsel-tax-audit-accounting ; https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/get-started/manage-your-tr-account.html ; https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/administration/metrics-page-admins-only ; https://www.thomsonreuters.com/en-us/posts/innovation/cocounsel-achieves-iso-iec-420012023-certification-another-milestone-in-trusted-ai/ ; https://ir.thomsonreuters.com/news-releases/news-release-details/thomson-reuters-reports-fourth-quarter-and-full-year-2025 ; US framework reference: https://www.nist.gov/itl/ai-risk-management-framework · Confidence: Low
What to know before you buy
These don't change the score above — they're conditions worth understanding for specific readers before you commit.
The vendor instructs users to review responses and citations carefully, while tax practitioners report that CoCounsel can still hallucinate: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/workspaces/manage-your-workspace and https://www.reddit.com/r/taxpros/comments/1uf6936/cocounsel_and_ready_to_advise/
Who this matters to: Firms seeking autonomous tax conclusions without qualified human review — An unchecked response could support an incorrect tax position, client communication, or workpaper conclusion despite containing citations.
What to do: Require qualified review, citation checking, documented approval, and escalation rules before outputs enter client work.
Changes our recommendation for this audience — see verdict notes.
The vendor says prompts and content are not used for GenAI training, but interactions may improve the product and account content remains until manually deleted: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/security/security-info-cocounsel-tax-audit-accounting
Who this matters to: US tax firms uploading returns, workpapers, or other sensitive client records — Client material can persist for the account lifetime, and the meaning of product-improvement processing may not satisfy every firm's data-governance policy.
What to do: Review the DPA and engagement terms, clarify improvement-use controls, minimize uploaded data, and establish retention and deletion procedures.
Vendor documentation advises exporting relevant conversations or workspaces before deleting a user and documents Excel workspace exports, but no comprehensive whole-account bulk export was confirmed: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/administration/view-users-roles-and-status.html
Who this matters to: Firms requiring complete, efficient data portability at contract end — Offboarding may require manual exports, and records could be lost if users are deleted before content is preserved.
What to do: Request written offboarding documentation, supported export formats, retention timing, and a test export before signing.
Vendor documentation describes administrative usage metrics and review statuses, but the reviewed public materials did not establish a detailed event-level audit log for research, document access, edits, and exports: https://www.thomsonreuters.com/en-us/help/cocounsel/tax-audit-accounting/administration/metrics-page-admins-only
Who this matters to: Accounting firms whose governance requires detailed access and activity evidence — The firm may be unable to reconstruct specific user actions or demonstrate its required oversight using standard product reports alone.
What to do: Obtain the security control matrix, audit-log field list, retention period, and sample export during procurement.
Recommendation depends on who's asking
Firms seeking autonomous tax conclusions without qualified human review: An unchecked response could support an incorrect tax position, client communication, or workpaper conclusion despite containing citations.
NTK Score: TaxGPT
Our editorial assessment of how confidently we'd recommend this to a small business, weighing real-world usability, commercial value, effort, long-term risk and practical experience — not a popularity score or a compliance audit.
The vendor states its US tax research provides primary-source citations, document analysis, and client context, strongly fitting tax firms, while practitioner signals and immature adjacent automation limit category maturity.
Tier 2 · TaxGPT product overview: https://www.taxgpt.com/what-is-taxgpt ; TaxGPT research documentation: https://support.taxgpt.com/en/articles/10108650-how-to-use-taxgpt-research ; practitioner discussion: https://www.reddit.com/r/taxpros/comments/1umrigq/taxgpt_any_thoughts/ · Confidence: Moderate
The vendor documents self-guided onboarding, email sign-in, CSV client imports, and simple exports, but deep desktop tax-software connectivity remains partly in development and migration guidance is incomplete.
Tier 2 · Getting-started collection: https://support.taxgpt.com/en/collections/9414238-getting-started ; account setup: https://support.taxgpt.com/en/articles/9355292-how-to-set-up-a-taxgpt-professional-account ; Cowork FAQ: https://support.taxgpt.com/en/articles/15216730-faq-s-taxgpt-cowork ; practitioner discussion: https://www.reddit.com/r/taxpros/comments/1umrigq/taxgpt_any_thoughts/ · Confidence: Moderate
Vendor documentation shows accessible plain-language workflows and extensive guidance, while practitioners report slow research, interface clutter, limited follow-up behavior, false positives, and continuing managerial supervision.
Tier 2 · TaxGPT help center: https://support.taxgpt.com/en/ ; client-profile guidance: https://support.taxgpt.com/en/articles/10120780-client-profiles-vs-general-chat ; G2 reviews: https://www.g2.com/products/taxgpt/reviews ; practitioner discussions: https://www.reddit.com/r/taxpros/comments/1umrigq/taxgpt_any_thoughts/ and https://www.reddit.com/r/taxpros/comments/1vhihjv/i_am_done_with_taxgpt_forever/ · Confidence: Moderate
Research-time savings are plausible but independently unbenchmarked, while quote-only procurement, undisclosed plan limits, and practitioner reports of roughly $1,600 to over $2,000 annually make SMB value difficult to verify.
Tier 2 · Current TaxGPT pricing page: https://www.taxgpt.com/pricing ; practitioner pricing and value discussion: https://www.reddit.com/r/taxpros/comments/1umrigq/taxgpt_any_thoughts/ ; recent billing allegation: https://www.reddit.com/r/taxpros/comments/1vhihjv/i_am_done_with_taxgpt_forever/ · Confidence: Moderate
Vendor-stated SOC 2 Type II, encryption, MFA, role controls, audit logs, US hosting, and CCPA terms are positives, but conflicting training and residency language plus mixed support signals weaken regulated-data confidence.
Tier 2 · Security page: https://www.taxgpt.com/ca/security ; pricing security claims: https://www.taxgpt.com/pricing ; privacy notice: https://www.taxgpt.com/legal/privacy ; DPA: https://www.taxgpt.com/legal/dpa ; funding announcement: https://www.taxgpt.com/blog/taxgpt-raises-4-6m-to-build-the-first-ai-tax-co-pilot-for-accounting-and-tax-firms ; G2 reviews: https://www.g2.com/products/taxgpt/reviews ; practitioner discussion: https://www.reddit.com/r/taxpros/comments/1vhihjv/i_am_done_with_taxgpt_forever/ · Confidence: Moderate
Trust score of 12/20 prevents a Recommended verdict. See Trust justification above.
What to know before you buy
These don't change the score above — they're conditions worth understanding for specific readers before you commit.
TaxGPT says it never trains on customer data at https://www.taxgpt.com/pricing, but its privacy notice permits personal-information use for model training at https://www.taxgpt.com/legal/privacy and DPA section 2.10 permits Customer Personal Data processing to improve service quality at https://www.taxgpt.com/legal/dpa.
Who this matters to: US tax firms uploading identifiable taxpayer records or correspondence — The documents do not establish an unambiguous no-training boundary, complicating the firm's Section 7216, FTC Safeguards Rule, confidentiality, and vendor-risk assessment.
What to do: Do not upload identifiable client data until TaxGPT provides written contractual clarification covering training, service improvement, human review, subprocessors, and opt-out rights.
Changes our recommendation for this audience — see verdict notes.
The vendor's security page says data is stored within the United States at https://www.taxgpt.com/ca/security, while its privacy notice says personal information may be processed and stored anywhere worldwide at https://www.taxgpt.com/legal/privacy.
Who this matters to: Firms with contractual or internal US-residency requirements — A firm cannot confidently map storage and processing locations or assess every subprocessor from the public documents alone.
What to do: Request the current data-flow diagram, subprocessor list, hosting regions, backup locations, and a binding US-residency commitment before procurement.
Recent practitioner posts allege Agent Andrew false positives, checkbox-reading failures, broken automated preparation, and unsupported desktop workflows: https://www.reddit.com/r/taxpros/comments/1vhihjv/i_am_done_with_taxgpt_forever/ and https://www.reddit.com/r/taxpros/comments/1umrigq/taxgpt_any_thoughts/ ; these are anecdotal Tier 3 reports, not independent benchmarks.
Who this matters to: Firms buying TaxGPT primarily for return review, tax preparation, or Cowork automation — Manager review and remediation could erase projected time savings or introduce workflow risk if local documents and tax software are not handled reliably.
What to do: Run a controlled trial using representative returns, checkboxes, scanned documents, and existing tax software; measure misses, false positives, completion rates, and manager review time.
Changes our recommendation for this audience — see verdict notes.
TaxGPT announced a $4.6 million funding round in February 2025 at https://www.taxgpt.com/blog/taxgpt-raises-4-6m-to-build-the-first-ai-tax-co-pilot-for-accounting-and-tax-firms, while 2026 help articles still describe several Cowork capabilities as forthcoming at https://support.taxgpt.com/en/articles/15216730-faq-s-taxgpt-cowork.
Who this matters to: Firms expecting multiyear platform stability or purchasing against roadmap commitments — Feature availability, integrations, packaging, and support commitments may change more quickly than with established tax-research vendors.
What to do: Contract against currently demonstrated functionality, document roadmap dependencies, and seek renewal and data-exit protections.
The help center documents exporting individual responses through email, copy, print, or PDF at https://support.taxgpt.com/en/articles/10119934-response-export-email-copy-print, but desk research found no public procedure for bulk exporting Client Intelligence records, documents, history, and metadata.
Who this matters to: Firms adopting Client Intelligence as a persistent client repository — Leaving the service could require manual extraction or result in loss of structured client context even though the DPA promises deletion after termination.
What to do: Require a demonstration and written specification of complete bulk export formats, metadata coverage, timing, cost, and post-termination access.
Recommendation depends on who's asking
US tax firms uploading identifiable taxpayer records or correspondence: The documents do not establish an unambiguous no-training boundary, complicating the firm's Section 7216, FTC Safeguards Rule, confidentiality, and vendor-risk assessment.
Firms buying TaxGPT primarily for return review, tax preparation, or Cowork automation: Manager review and remediation could erase projected time savings or introduce workflow risk if local documents and tax software are not handled reliably.
NTK Score: CCH AnswerConnect
Our editorial assessment of how confidently we'd recommend this to a small business, weighing real-world usability, commercial value, effort, long-term risk and practical experience — not a popularity score or a compliance audit.
The vendor states that federal and multistate research, primary sources, cited AI answers, document analysis, and CCH integrations fit daily US tax work, but specialty depth costs more and practitioner reports on search and AI quality are mixed.
Tier 2 · https://www.wolterskluwer.com/en/solutions/cch-answerconnect-us ; https://shoptax.wolterskluwer.com/en/cch-answerconnect-tax.html ; https://www.reddit.com/r/taxpros/comments/1ow3wrs/cch_users_how_are_you_liking_the_ai_feature_for/ · Confidence: Moderate
Cloud access, a one-week trial, guided walkthroughs, included webinars, and direct CCH-suite links reduce setup, but benefits narrow outside CCH workflows and replacing established libraries, bookmarks, and staff habits still requires migration work.
Tier 2 · https://www.wolterskluwer.com/en/solutions/cch-answerconnect-us ; https://cch-answerconnect.groovehq.com/help/getting-started-60b09170 ; https://cch-answerconnect.groovehq.com/help/how-to-do-tax-research-training ; https://www.reddit.com/r/taxpros/comments/1getdtg/tax_and_accounting_research_software/ · Confidence: Moderate
Natural-language chat, citations, consolidated topic views, and complimentary training support adoption, but practitioner signals range from useful to clunky, hard to search, and weaker than rival AI, implying uneven training needs.
Tier 2 · https://cch-answerconnect.groovehq.com/help/cch-answerconnect-new-features-december-2024 ; https://cch-answerconnect.groovehq.com/help/guided-walkthroughs ; https://cch-answerconnect.groovehq.com/help/how-to-do-tax-research-training ; https://www.reddit.com/r/taxpros/comments/1ow3wrs/cch_users_how_are_you_liking_the_ai_feature_for/ ; https://www.reddit.com/r/taxpros/comments/1getdtg/tax_and_accounting_research_software/ · Confidence: Moderate
A published $1,499 per-person entry price and potential research-time savings support ROI for active tax teams, but billing term, seat discounts, renewals, specialty tiers, and expanded-package pricing remain unclear, while practitioners report materially higher annual totals.
Tier 2 · https://shoptax.wolterskluwer.com/en/cch-answerconnect-tax.html ; https://www.wolterskluwer.com/en/solutions/cch-answerconnect-us/packages ; https://www.reddit.com/r/taxpros/comments/1hrvlbj/thinking_of_dumping_cch_answerconnect_and_going/ ; https://www.reddit.com/r/taxpros/comments/1o98zd0/cocounsel_checkpoint_edge_vs_bluej_vs_cch/ · Confidence: Moderate
Wolters Kluwer states that customer data is not used to train AI and describes encryption and independent audits, but AnswerConnect-specific SOC 2 scope, residency, document retention, MFA, and comprehensive audit controls were not publicly verified for FTC Safeguards Rule due diligence.
Tier 2 · https://www.wolterskluwer.com/en/about-us/artificial-intelligence ; https://assets.contenthub.wolterskluwer.com/api/public/content/9f3a75d5d13444adbbe27bccab7f468a ; https://cch-answerconnect.groovehq.com/help/managing-users ; https://www.ftc.gov/business-guidance/resources/ftc-safeguards-rule-what-your-business-need-know ; https://www.reddit.com/r/taxpros/comments/1getdtg/tax_and_accounting_research_software/ · Confidence: Moderate
Trust score of 13/20 prevents a Recommended verdict. See Trust justification above.
What to know before you buy
These don't change the score above — they're conditions worth understanding for specific readers before you commit.
CCH online terms describe encryption and SOC 2 Type 2 audits for select systems, but public AnswerConnect materials do not identify whether this product is in scope or document its hosting region, uploaded-file retention, deletion, MFA, or subprocessors: https://assets.contenthub.wolterskluwer.com/api/public/content/9f3a75d5d13444adbbe27bccab7f468a ; FTC service-provider guidance: https://www.ftc.gov/business-guidance/resources/ftc-safeguards-rule-what-your-business-need-know
Who this matters to: US tax firms uploading identifiable client returns or requiring documented vendor controls — The firm may be unable to complete service-provider due diligence or align access and retention controls with its written information security program.
What to do: Obtain the current SOC 2 Type 2 report and bridge letter, DPA, subprocessor list, hosting locations, MFA or SSO documentation, audit-log details, and uploaded-file retention and deletion terms.
Changes our recommendation for this audience — see verdict notes.
Current marketing describes Expert AI answers as validated by in-house tax experts, while the vendor's May 2024 help article warns that GenAI answers are not expert-validated and are not always reliable: https://www.wolterskluwer.com/en/solutions/cch-answerconnect-us ; https://cch-answerconnect.groovehq.com/help/ai-enabled-capabilities
Who this matters to: Firms expecting AI answers to replace source-level professional review — Treating every generated response as individually expert-approved could produce unsupported tax advice when the underlying source and reasoning have not been checked.
What to do: Ask the vendor which current outputs receive human validation, test representative complex questions during the trial, and require staff to verify cited primary sources before client use.
Changes our recommendation for this audience — see verdict notes.
Recommendation depends on who's asking
US tax firms uploading identifiable client returns or requiring documented vendor controls: The firm may be unable to complete service-provider due diligence or align access and retention controls with its written information security program.
Firms expecting AI answers to replace source-level professional review: Treating every generated response as individually expert-approved could produce unsupported tax advice when the underlying source and reasoning have not been checked.
NTK Score: Bloomberg Tax AI Assistant
Our editorial assessment of how confidently we'd recommend this to a small business, weighing real-world usability, commercial value, effort, long-term risk and practical experience — not a popularity score or a compliance audit.
Purpose-built tax dialogue, linked source material and Bloomberg’s extensive practitioner content strongly fit complex US tax research, although drafting, client-document analysis and independent accuracy benchmarking remain limited or unverified.
Tier 2 · Tier 2 vendor materials: https://pro.bloombergtax.com/products/ai-and-bloomberg-tax/ and https://pro.bloombergtax.com/insights/tax-automation/ai-for-tax-professionals/ ; Tier 3 corroboration: https://tax.thomsonreuters.com/blog/how-to-choose-the-best-ai-tax-research-tool/ and https://www.capterra.com/p/10036786/Bloomberg-Tax-for-Small-CPAs/ · Confidence: Moderate
Natural-language research and platform-wide availability should reduce initial technical work for existing subscribers, but newcomers face sales-led onboarding, an established research interface and poorly documented third-party integrations or migration paths.
Tier 2 · Tier 2 vendor materials: https://pro.bloombergtax.com/products/ai-and-bloomberg-tax/ and https://pro.bloombergtax.com/insights/company-news/bloomberg-tax-accounting-introduces-innovation-studio-development-approach/ ; Tier 3 signals: https://www.gartner.com/reviews/product/bloomberg-tax?marketSeoName=tax-software and https://www.capterra.com/p/10036786/Bloomberg-Tax-for-Small-CPAs/ · Confidence: Moderate
Everyday-language questions, linked references, training resources and advertised round-the-clock support support adoption, while sparse Assistant-specific reviews and generic Bloomberg Tax feedback indicate some initial complexity and usability friction.
Tier 2 · Tier 2 vendor pages and support guide: https://pro.bloombergtax.com/products/ai-and-bloomberg-tax/ and https://btax-help.bloombergtax.com/images/50-9705-22409-tm-btax-qrc-bravo.pdf ; Tier 3 signals: https://www.gartner.com/reviews/product/bloomberg-tax?marketSeoName=tax-software and https://www.capterra.com/p/10036786/Bloomberg-Tax-for-Small-CPAs/ · Confidence: Moderate
Frequent complex research could justify the subscription through saved professional time, but Bloomberg publishes no clear official price and current third-party listings conflict over whether the reported $2,200 starting figure is monthly or annual.
Tier 2 · Tier 2 vendor page requiring a demo request: https://pro.bloombergtax.com/products/ai-and-bloomberg-tax/ ; conflicting Tier 3 pricing sources: https://www.capterra.com/p/10036786/Bloomberg-Tax-for-Small-CPAs/ and https://www.bizora.ai/post/how-much-should-cpas-pay-for-ai-tax-research-tools · Confidence: Low
The vendor states SOC 2, CCPA/CPRA processor limits, audit logging and role-based controls, and Bloomberg is stable; deductions reflect unclear AI-specific certification scope, prompt retention, model-training use, residency and sparse Assistant-specific support evidence.
Tier 2 · Tier 2 vendor trust materials and Bloomberg Tax Research DPA: https://trust.bloombergindustry.com/ and https://assets.bbhub.io/bna/sites/10/2023/12/2023_Data-Processing-Agreements_Bloomberg-Tax-Research.pdf ; Tier 3 support signal: https://www.gartner.com/reviews/product/bloomberg-tax?marketSeoName=tax-software · Confidence: Moderate
What to know before you buy
These don't change the score above — they're conditions worth understanding for specific readers before you commit.
The 2023 Bloomberg Tax Research DPA limits processing to customer instructions and provides security and deletion provisions, but reviewed public material did not clearly establish current AI prompt retention, training use, model providers or residency: https://assets.bbhub.io/bna/sites/10/2023/12/2023_Data-Processing-Agreements_Bloomberg-Tax-Research.pdf
Who this matters to: US tax firms intending to enter taxpayer PII, client documents or confidential fact patterns into AI Assistant — Without Assistant-specific contractual terms, a firm cannot confidently determine whether its intended data use matches internal confidentiality controls, CCPA/CPRA expectations or its NIST-aligned AI risk process.
What to do: Obtain the current AI FAQ, DPA, subprocessor list, SOC 2 scope and written prompt-retention and model-training terms before permitting client-identifiable information.
Reviewed vendor materials did not document an export route for Assistant conversations or research history, while the small Gartner review sample includes a generic Bloomberg Tax vendor-lock-in concern: https://www.gartner.com/reviews/product/bloomberg-tax?marketSeoName=tax-software
Who this matters to: Firms that need portable research records or expect to change research platforms — Saved research context may require manual preservation, increasing switching effort and weakening continuity if the subscription ends.
What to do: Ask Bloomberg to demonstrate conversation, citation and workspace export, retention after cancellation, and administrator-controlled deletion before contracting.